GSoC 2022 call | Catalyst

This is a list of projects suggested by Catalyst-Team for GSoC 2022. GSoC has focused on 2 types of projects this year: ~175 hours or ~350 hours to complete, the list below contains project suggestions that should meet these criteria depending also on the skill level of the participant. These are only suggestions, and if you have your own ideas, please discuss them on the Catalyst Team Slack. We have a lot of talented ML practitioners in the Catalyst Team who would love to mentor good students for GSoC 2022.
See lower down on this page for more details for some of the projects listed above.
The timeline for GSoC 2022 is here.
When making the difficult decision about which students to accept, we look for:
Like every open-source framework, Catalyst is actively interested in examples and documentation improvements to improve its adoption. In such a case during any listed projects below you are welcome to extend our documentation site with FAQ of your own, or Keras-like examples.
LR finder is a well-known technique in the ML community to tune your initial learning rate for maximal performance. While Catalyst already has its implementation with some extensions there is a place for improvement in terms of user API.
The goal of the project
The main idea of this project is to transfer Catalyst from pip to a poetry-based development environment. It includes working with installations scripts, tests, and contributions guides. See RP#1358 for more information.
Catalyst Team was always inspired with the Metric Learning. For such a reason, we also have a wide variety of ML-based metrics: CMCMetric, ReidCMCMetric.
The goal of this project
In addition to Metric Learning, Catalyst also has a Self-Supervised Learning benchmark. Nevertheless, it could be improved a bit.
The goal of this project
There are many RecSys-based examples in the Catalyst already. One thing to mention - they are mainly AutoEncoders-based. For such a reason we are interested in its adoption for a wide variety of sequential recommenders also.
The goal of this project
While Catalyst already has a few off-policy RL examples, several improvements can be made.
The goal of this project
While Catalyst already has a few on-policy RL examples, several improvements can be made.
The goal of this project